Integration of spectral indices and precipitation data to assess river morphometric features in a tropical semi-humid environment

• Workflow for river centreline mobility and geometrical measurements. • The random forest algorithm achieved water masking accuracy exceeding 80%. • Quantification of abrupt river adjustments in data-scarce contexts. • Hotspots of heterogeneity in local morphodynamics on low-gradient valleys. • The results represent an initial examination of the river path at the reach scale. In mixed fluvial corridors, calculating a mobility zone is important due to the geometric instability caused by abrupt channel changes. Thus, this work aims to develop a reproducible remote sensing workflow using eight optical indices and a random forest algorithm to evaluate river channel mobility on a mixed-pattern river in data-limited fluvial contexts. This work used 13 images from 1985 to 2024 to extract the river centerline for the quantification of lateral shift and sinuosity, and to examine the relationship with hydrological forcing. The algorithm achieved 87% accuracy, a beneficial result for water mask extraction. Lateral shift revealed reach-specific high-magnitude adjustment events (up to ∼500m) concentrated in localized river segments. Sinuosity showed a persistent curvilinear planform (>1.5), and changes in different dates were minor compared to lateral shift variability. Time-lag correlation indicated that hydrological forcing influences centerline changes at short temporal lags (r = -0.804, p = 0.029), with no evidence of river memory effects. In general, the results captured episodic and reach channel adjustments, presenting a practical basis for monitoring channel mobility and fluvial exposure hazards in the absence of detailed information.